1 / 55

Physical layer network coding for next generation wireless broadband

Physical layer network coding for next generation wireless broadband. Alister Burr, University of York agb1@ohm.york.ac.uk Agisilaos Papadogiannis, Chalmers University. Outline. The challenge of next generation wireless networks Next generation network architectures MIMO and MIMO cellular

bena
Download Presentation

Physical layer network coding for next generation wireless broadband

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Physical layer network coding for next generation wireless broadband Alister Burr, University of York agb1@ohm.york.ac.uk Agisilaos Papadogiannis, Chalmers University

  2. Outline • The challenge of next generation wireless networks • Next generation network architectures • MIMO and MIMO cellular • Multi-user and Network MIMO • Physical layer network coding • Conclusions

  3. Wireless networks – some history • Began with Marconi in 1890’s • … • 1st generation • analogue, telephony ~1980 (Japan) • 2nd generation • digital, some data ~1992 • 3rd generation • CDMA, flexible services, up to 384 kbit/s ~2002 • 4th generation • OFDM(A), full Internet access, up to 1 Gbit/s ???

  4. 4G • The “next generation” has been discussed ever since 3G standards were finalised a decade ago • however it was not initially clear what form the “fourth generation” might take • However starting from 2002 ITU-R has defined the requirements for IMT-Advanced • which has since then been generally accepted as the definition of 4G • Key requirement is 100 Mbit/s for high mobility and 1 Gbit/s for low mobility • Standards currently under development: • Mobile WiMAX (IEEE 802.16e/m) • 3GPP LTE-Advanced

  5. The dream • To provide full Internet connectivity to everyone, anywhere • which means wirelessly • Next generation wireless research has usually focussed on a ‘headline’ maximum data rate • but of course this will not be the rate most users experience, • and probably is not the most important figure • In densely-populated cities a network for everyone must provide extremely high capacity densities

  6. Required capacity density • Average population density in European cities ranges from 3400 - 5400/km2 • however in commercial district in working hours it will be much higher • say 8000/km2 • Suppose 10% subscribe, and 20% of those require access at busy hour • Expected data rate 5 Mbit/s 8000/km2 10%  20%  5 Mbit/s = 800 Mbit/s/km2

  7. Capacity density of 1G • e.g. AMPS, U.S.A: • ~400 channels in each direction • ~15 km radius cells  700 km2 • re-use ~1:10  0.06 channels/km2, • equivalent to approx 1.8 kbit/s/km2 • in approx 50 MHz

  8. Current and 4G systems • Currently one base station serves about 1km2 • 4G bandwidths proposed are ~ 40 MHz • Best available bandwidth efficiency averages about 2 bits/s/Hz across cell • hence capacity density is 80 Mbit/s/km2 • assumes 100% frequency re-use • We need an order of magnitude more! • 10 more bandwidth unlikely to be available

  9. Increased density • We can also increase number of cells • BUT • need many more cell sites • interference

  10. Cell edge problem • We may be able to increase bandwidth efficiency (bits/s/Hz) • use (e.g) advanced MIMO techniques • BUT mobile close to cell edge suffers interference from adjacent cells • Conventionally we reduce frequency re-use • but this reduces available bandwidth by factor 3 or more

  11. BuNGee • University of York is part of a European project tackling these problems • Beyond Next Generation Mobile Broadband (BuNGee) • Proposes: • new hierarchical network architecture based on wireless backhaul • Advanced MIMO techniques for high bandwidth efficiency • Self-organising network for optimal spectrum use • Goal: 1 Gbit/s/km2

  12. Outline • The challenge of next generation wireless networks • Next generation network architectures • MIMO and MIMO cellular • Multi-user and Network MIMO • Physical layer network coding • Conclusions

  13. Achieving capacity density • Will probably need a combination of the approaches mentioned: • More spectrum • Improved bandwidth efficiency • especially increased use of MIMO • Increased frequency re-use (100%) • Reduced cell size • requires low base station installation cost, • and a cost-effective backhaul network

  14. Wireless backhaul • Simple comparison with 4G proposals suggests we may need ~10 BSs per km2! • We believe that the only cost-effective way to provide this is by wireless backhaul • However must allow for spectrum used by backhaul links • Hence must minimise backhaul load

  15. BuNGee architecture

  16. Some figures • Assume HBS serves 1 km2 • Assume total 40 MHz available • 20 MHz for MS-ABS (access links); • 20 MHz ABS-HBS (backhaul) • Assume average 2 bits/s/Hz across cell • Then capacity per ABS = 20  2 = 40 Mbit/s • No. ABS per HBS = 1 Gbit/s / 40 Mbit/s = 25 • Area served by ABS = 1 km2/25 = 40 000 m2, or 200m square

  17. ABS MS ABS MS HBS Relay ABS Wireless mesh network • Since cells are very small, • mobile (MS) may be served by more than one ABS • MSs now served by optimum combination of available ABSs • Practically abolishes concept of cells! • Overall network looks more like a wireless mesh network

  18. Wireless relaying • A cell can be extended by adding fixed, or infrastructure relays • very similar architecture to wireless backhaul • with relays replacing ABSs • may allow direct connection of MSs to hub Relay Relay

  19. Hierarchical wireless network • A generalised framework for network architectures involving wireless backhaul and/or relaying • we might allow: • more than one layer of relays • direct connections between nodes on the same level • MSs to connect to different relay levels • Again, similar to mesh network in structure

  20. Outline • The challenge of next generation wireless networks • Next generation network architectures • MIMO and MIMO cellular • Multi-user and Network MIMO • Physical layer network coding • Conclusions

  21. H11 ENCODER DECODER H21   HnR1 r s MIMO link model • MIMO = Multiple Input, Multiple Output • i.e. multiple antennas at each end of a link • Input and output signal can be modelled as (1  nT) and (1  nR) vectors, s and r; noise (1  nR):n • channel modelled as a matrix H:element Hij gives propagation between transmit antenna j and receive antenna i r = H s + n

  22. n 1 n H s r' s' U VH r n nR nT Eigendecomposition • Multiply by matrices U and V at input and output of channel, where • columns of U,V are transmit and receive eigenvectors of the channel • Then U H VH =  • a diagonal matrix with the square roots of the eigenvalues of HHH on the diagonal, and r’i = is’i + n’i • i.e. we create a set of uncoupled channels, whose power gains are the eigenvalues • Each eigenvector can be treated as a steering vector for antenna array  transmit/receive beam patterns

  23. Beamforming model n • Another way of viewing MIMO: • each input stream corresponds to a beam from the antenna array towards a multipath signal • can create as many such beams as there are antennas • hence can transmit up to n = min(nT, nR) beams • provided there are enough multipaths r' s' U VH

  24. MIMO capacity • Capacity and bandwidth efficiency approximately multiplied by no. of streams, n • Slope of curves proportional to n • called multiplexing gain • Dramatic capacity gain!

  25. U int d s int int Beamformer H int W V ˆ d r x U H matched d s n filter MIMO in interference • MIMO cellular system also subject to interference • Beamformer applies linear weights to maximise SNIR at output • filters signals from different directions to maximise signal to noise-plus-interference ratio

  26. Capacity of MIMO cellular • 4 antenna elements • MIMO system capacity around 3  SISO • and more than 1.5  SIMO (smart antenna) • Beamformer (“prewhitening”) very important • Interference limited

  27. Implications for 4G networks • MIMO can dramatically increase link capacity • and significantly increase cellular capacity • Note that capacity is mainly affected by n = min(nT, nR) • Still severely limited by inter-cell interference • Can we reduce the effect of interference?

  28. Outline • The challenge of next generation wireless networks • Next generation network architectures • MIMO and MIMO cellular • Multi-user and Network MIMO • Physical layer network coding • Conclusions

  29. Multi-user MIMO systems • It has been known since the 1960s that the optimum means of sharing a channel between several users may be by simultaneous, mutually interfering transmission • as opposed to time-division multiplexing, or other orthogonal multiplexing • Information-theoretic approach: • Multiple Access Channel - MAC (uplink) • Broadcast Channel - BC (downlink)

  30. R2 C2 time-sharing rate Csum C1 R1 2-user SISO MAC • Rate region: set of achievable rates of the two sources • C1 and C2 are capacities of two channels without the other • “Corner points” achieved by successive interference cancellation • “Time sharing” sum rate limited to dashed line • in general achievable sum rate Csumexceeds this

  31. Multi-user MIMO MAC • In a multi-user MIMO multiple access channel (uplink), • sum rate capacity limit is capacity of MIMO channel formed by combining all Tx antennas of all users • For nU users with nT Tx antennas, average Tx power Si and Tx time Ti(each) • where: • denotes capacity of nTnR MIMO channel with SNR S

  32. nT BS nU nR Multi-user MIMO • Conventional TDMA/FDMA is equivalent to time-sharing • divides “headline” rate by no. of channels • MU-MIMO allows several users to share same time slot/channel • Users/BS can act as a single nT nU  nR MIMO system • Usually more BS than terminal antennas • multiplexing gain no longer limited by no. terminal antennas

  33. Symmetric and asymmetric • Modest advantage when nT = nR (symmetric links) • Large advantage when nT nR (asymmetric links) • max. multiplexing gain becomes min(nR, nTnU)

  34. R2 C2 time-sharing rate Csum C1 R1 Broadcast channel • Broadcast channel simply means one transmitter to many receivers • obvious example is radio/TV broadcasting, where same message is intended for all • also applies to cellular downlink, where a different message is intended for each receiver • In latter case can define a capacity region like that for multi-user MIMO • Again time sharing (TDMA/FDMA) is sub-optimal

  35. Dirty paper coding (DPC) • This is an information-theoretic result which applies to a channel subject to interference where the interference is known at the transmitter • r = s + i + n • where s is the information signal, i is the interference, and n is noise • Achievable capacity of this channel is log2(1 + PS/Pn) • i.e. the same capacity as if the interference were not present • note this is not achieved by “pre-cancelling” the interference • On a broadcast downlink, the signal to one user is interference to another, and is known at the transmitter

  36. f(x) x A Precoding in practice • Dirty paper precoding in principle operates by selecting a codebook (set of transmitted codewords) depending on the interference • A more practical scheme following the same principle is Tomlinson-Harashima precoding (THP) • Use modulo function f(x), and transmit: • where d is data, i is interference and k is some integer • At the receiver apply the modulo operation again: • interference is removed • some degradation due to “folding” of noise

  37. Linear beamforming • Or simpler still, we can simply form beams to each user • ensuring also that we null interference to other users • this is a purely linear operation

  38. Network MIMO • So far we have assumed that signals from other cells must be treated as interference • However it is possible for several (in principle, all) base stations to cooperate to transmit to a given mobile • or to receive from that mobile • Then there is in principle no CCI! • since all received signals are exploited as signals • The entire system then operates as a multi-user MIMO system with (on the uplink) nTnU nC transmit and nRnC receive antennas • where nC is the number of cooperating cells • in principle multiplexing gain approaches min(nRnC, nTnU nC)

  39. Practical limitations • There are of course practical limitations to this concept: • Where should the processing be performed? • In a hierarchical network like BuNGee, at the hub • Distributed methods also possible, with processing at cooperating BSs • Computational complexity • Synchronisation • Is it feasible to keep cooperating base stations phase synchronous? (especially for downlink) • Backhaul capacity

  40. Backhaul capacity requirements • On the downlink, if two BSs cooperate to communicate with an MS, that MS’s data should be sent to both • could double backhaul requirements • On the uplink, neither may be able to decode the MS without the signal from the other • hence analogue signal may need to be transmitted over the backhaul in high precision • may increase backhaul requirements by several times • Need to ensure backhaul links are efficiently used b1s1 b2s2 r2 r1 s1 s2 a1r1 + a2r2

  41. Limitations of in-band backhauling • If we use wireless backhaul, we must account for bandwidth occupied • Can we re-use the same spectrum in backhaul and access segments? • in-band backhauling • Duplexing restrictions of ABSs usually prevent same resources being used in the two segments in the same place

  42. Outline • The challenge of next generation wireless networks • Next generation network architectures • MIMO and MIMO cellular • Multi-user and Network MIMO • Physical layer network coding • Conclusions

  43. Network coding • A network node applies a joint coding function to two (or more) incoming data streams • instead of simply switching between them • In this simple example (the “butterfly network”) the central node applies the XOR function (modulo-2 addition) • then both streams can be recovered at both output nodes • without network coding the central link would have to have twice the capacity to achieve this

  44. SA b SB a R DB DA a  b a  b Two-way relay channel (2WRC) • Allows a relay to support transmissions in two directions at once • Relay broadcasts XOR combination of two incoming streams • Each destination can then reconstruct data intended for it by XOR combination with the data it transmitted

  45. Physical layer network coding (PLNC) • In a wireless network, we do not have discrete, non-interfering paths • except by using TDMA or FDMA • Signals: • are broadcast to all nodes within range • combine additively in signal space • However it is still possible to extract a joint information stream equivalent to XOR combination

  46. a b SA SB a + b DB R DA a  b a  b a b a+b a  b 0 0 -2 0 0 1 0 1 1 0 0 1 1 1 +2 0 PLNC for 2WRC • System operates in two phases • Phase 1: sources transmit simultaneously • Phase 2: relay transmits • Assume both sources transmit BPSK • {+1, -1}  {1, 0} Phase 1 Phase 2 SA, SB SA, SB time

  47. SA b SB a R DB DA a  b a  b SA SB R(A) R(B) time SA SB SA, SB R(AB) R(AB) time time For comparison • Without network coding • Network layer network coding • Physical layer network coding

  48. R S D Relay data compression • Cooperative diversity: relay provides extra diversity if S – D link fades • Phase 1: Source transmits to relay and destination • Phase 2: Relay transmits to destination • Note signal at relay is correlated with Phase 1 signal at destination • since both arise from the same data • This allows distributed compression using Slepian-Wolf coding • reduces the data relay must transmit

  49. Slepian-Wolf coding R1 • If two data sources are correlated, their joint information content is less than the sum of their separate content • Can exploit this to compress the data • even though encoders are separate • White area on graph gives rate region • range of possible compressed rates to allow reconstruction S’1 S1 C1 D R2 S’2 S2 C2 R2 R1

  50. T1 T2 B1 B2 B3 Hub PLNC in network MIMO • Example: • 2 terminals connectedto hub via 3 BS • B2 can use PLNC, to shareits link with the hub betweentwo terminals • It can use distributed compression, since the data is correlated with that via B1 and B3 • Reduces backhaul load

More Related